Developing a long-term live-fuel moisture content dataset based on passive microwave vegetation optical depth

Abstract

Live fuel moisture content (LFMC) describes the water content of the living vegetation (leaves, grass) relative to the dry matter content. LFMC is an important feature for the description of the vegetation status and often used in the description and prediction of fire ignitions and spread. LFMC has been widely estimated from medium resolution optical satellite imagery. However, passive microwave satellite observations are also sensitive to LFMC. The long temporal coverage and almost daily availability of these observations allows to observe vegetation properties over long time scales at high temporal resolution but broad scale. Vegetation optical depth (VOD) from passive microwave sensors describes the attenuation of the microwave emission due to the vegetation layer. VOD depends on the vegetation water content (VWC), the mass of vegetation and the structure of the vegetation layer. VWC is a function of above-ground biomass and LFMC and hence LFMC can be potentially estimated from VOD. Here we present a methodology to estimate a new large-scale LFMC dataset from passive microwave VOD and leaf area index (LAI) by using long-term satellite records since 1987. In this study several model approaches describing the relation between LFMC, VOD and LAI were developed. These models were implemented and tested with three harmonised VOD datasets differing by used wavelengths for retrieval (Ku-band, X-band and C-band) of the vegetation optical depth climate archive (VODCA version 1) as well as with MODIS-LAI product MOD15A2H v006. For model calibration measurements of LFMC obtained from a global database of on-ground destructive LFMC observations (Globe-LFMC database) were utilised. The models were evaluated using spatial cross-validation with Globe-LFMC data as reference and afterwards compared with a MODIS-derived LFMC dataset of Australia and Europe. The evaluation were also conducted for specific vegetation forms. Across all models, the lowest RMSE and highest correlation between in-situ measured and derived LFMC were obtained for grasses, shrublands and broad-leaved deciduous trees. Lower performances were obtained for needle-leaved and broad-leaved evergreen trees. The best performing model approach (correlation of 0.4-0.8 and RMSE of 20%-120% LFMC depending on vegetation type) using Ku-band VOD as input was selected to estimate LFMC for all vegetation types at large scale. Large-scale global patterns of LFMC and the related uncertainties match expected spatial patterns and temporal dynamics. For example, anomalies in LFMC correspond well to known regional drought events. In order to extend the LFMC for the years before 2000, a new harmonised long-term VOD dataset (VODCA version 2) and several long-term LAI datasets were used as alternative input for the best performing model, specifically multisensory LAI from SPOT/PROBA-V/Sentinel-3 and GIMMS. The distinct derived LFMC datasets differing by the used LAI dataset as input were evaluated with global cross validation and compared with other LFMC datasets. These models provide a comparable performance to the original MODIS-based model and enables the estimation of long-term changes in LFMC at global scale. The use of new Sentinel-based datasets allows continuing this LFMC data record into the future which will be beneficial for analyses of vegetation water status and wild fire.